HINoV: A New Model to Improve Market Segment Definition by Identifying Noisy Variables

Abstract

Although cluster analysis is the procedure most frequently used to define data-based market segments, it is not without problems. This research addresses one of its major problems: the selection of the "best" subset of variables on which to cluster. If this selection is not made carefully, "noisy" variables that contain little clustering information can cause misleading results. To help isolate potentially noisy variables prior to clustering, the authors discuss a new algorithm, the Heuristic Identification of Noisy Variables (HINoV). They demonstrate its robustness with artificial data. In addition, the authors illustrate the potential of HINoV to yield more managerially useful market segments (clusters) when applied to two real marketing data sets. Implementation of HINoV is straightforward and will help avoid a major problem in using K-means cluster analysis for market segment definition, as well as for other similar types of research.

Selection of Variables in Cluster Analysis: An Empirical Comparison of Eight Procedures

“…Method 6: HINoV (Carmone et al, 1999) The HINoV clustering procedure developed by Carmone et al (1999) consists of the following steps:…”

“…Although Carmone et al (1999) recommend ranking the TOPRI(v) values in descending order and using a scree-type plot to determine the cutoff point for selected variables, this procedure is not computationally feasible for a large experimental study. Therefore, we formalized Carmone et al's method by using the ratio rule in Step 5.…”

“…These procedures have been developed in both the context of model-based clustering (Dy & Brodley, 2004;Law, Figueiredo, & Jain, 2004;Raftery & Dean, 2006) and non-model-based clustering (Brusco & Cradit, 2001;Carmone, Kara, & Maxwell, 1999;Friedman & Meulman, 2004;Montanari & Lizzani, 2001). Excluding the work of Brusco and Cradit (2001) and Carmone et al (1999), when new procedures are introduced they are normally demonstrated on a few "choice" data sets and comprehensive comparisons are never provided. Unfortunately, introducing new variable selection procedures in this manner results in an entire collection of techniques where there are no definitive recommendations about when to use which procedure.…”

Amalgamation of partitions from multiple segmentation bases: A comparison of non-model-based and model-based methods

“…We selected Hubert and Arabie's (1985) adjusted Rand index as the measure of agreement between the consensus partition and the validation partitions. This decision was based on the well-recognized properties of the ARI in the classification literature (Steinley, 2004), as well as its importance in market segmentation research (Helsen and Green, 1991;Carmone et al, 1999). Simulation studies have shown that ARI is the most desirable index for measuring segment recovery (Steinley, 2004).…”

Covariance-Guided Mixture Probabilistic Principal Component Analysis (C-MPPCA)

“…That said, DBSCAN is one of many fast, algorithmic, clustering methods. We considered a few to initialize C-MPPCA and formally compare DBSCAN to the one of the best methods reported in Steinley and Brusco (2008), heuristic identification of noisy variable (HINoV) (Carmone, Kara, and Maxwell 1999). As summarized in Section 5, DBSCAN outperforms HINoV.…”